What does it mean for two transactional applications to be similar? We address this question in this paper by highlighting four distinctive features of transactional applications: (1) the transaction size, i.e., the average number of memory accesses of the transactions; (2) the read-write ratio, i.e., the ratio between the number of accesses that modify the data and those that do not; (3) the contention, i.e., the number of concurrent accesses to the same shared data, such that at least one of these accesses is a write; (4) the uniformity, i.e., the extend to which transactions access distinct objects. We show that the similarity between an application A and an application Aʹ can be derived from these features and can be used to determine which concurrency control implementation works best for A based on having tested which worked best for Aʹ. We convey the accuracy of the profiling and predictions based on a study with six workloads and ten concurrency control mechanisms.
CITATION STYLE
Gramoli, V., Guerraoui, R., & Kermarrec, A. M. (2015). Profiling transactional applications. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9466, pp. 278–292). Springer Verlag. https://doi.org/10.1007/978-3-319-26850-7_19
Mendeley helps you to discover research relevant for your work.